I don't know either. Is it easier to confuse one's identity in ML as to whether one is doing computer science, computational science, application development, or even consulting/services?

There is obviously a long history of "resource disparity" in high-performance and distributed computing worlds. When I was more enmeshed in that field in the late 90s-early 2000s, I did not see this raising too much concern among my peers. You had the full gamut of domain scientists with interest in simulation to tool/framework builders and consulting specialists in computation. You could see collaborations publish domain results (e.g. geophysics or high-energy physics); CS/software results for parallel programming models, languages and libraries; more specific applied math results for the occasional novel numerical method; and of course EE/industrial results for the constantly churning hardware platforms put into service. Even then, there were of course "hero class" experiments which were notable more for scale than for actually being a new method or for producing a truly new insight in the application domain.

But, there were well established supercomputer centers with federal support and multiple ways for researchers to get machine time. Commercial players might have private resources, but were often paying clients who used government operated machines. For example, NASA sites would host private simulation runs for aerospace companies. It wasn't yet true (or at least not obviously true to the researcher at large) that private entities might have larger resources at their disposal than an academic with a typical grant. This may have allowed us all to imagine a more egalitarian field, where people believed they could win machine time grants in the same world-class systems.

Of course, there are/were always insiders with better access. Even a new supercomputer paid for by the government would usually be earmarked for the darlings of a particular program manager or committee to get early access and run wild on the machine to help field test it before it went into its planned production use. These often produced a set of papers that nobody else was going to be able to publish.